Page 1 of 1

electronic convergence problem on molecule

Posted: Sat Oct 13, 2012 8:39 am
by baliuzeger
I'm running calculations on organic molecule PTCDA by VASP 4.6,
and now I got problems on electronic convergence.

When I was doing ionic relaxation,
the electronic relaxation took 141 steps to accuracy of 10^(-4) for the 1st scf loop,
and all subsequent scf loops took only a few steps after movements of ions.
This calculation is done with 15A vacuum (15x20x30 unit cell) at each sides, Tetrahedron method with Blochl correction, 1x3x2 k-points, PAW-GGA PPs.

Code: Select all

!Job
SYSTEM=PTCDA
NWRITE=2
LELF=T

!Electronic
ENCUT= 520 eV
EDIFF= 1E-05
NELM=1000 
PREC = high
NBANDS=100
ISMEAR = -5
LORBIT = 12
ISTART=0
ICHARG= 0
LCHARG = TRUE
ISPIN=1

!Ion
ISIF=2
IBRION= 2
NSW = 1000
NBLOCK = 1
KBLOCK = 1
LCORR = T

!Computaion
NSIM = 2
NPAR = 1
LPLANE = TRUE
ALGO = Fast
LREAL = Auto


After the ionic relaxation, I started a series of test for convergence of Energy with respect to ENCUT (from 300 to 520),
then I got convergence problems on electronic relaxations:
only the one with ENCUT=440 converged in 187 steps and all others cannot converge in 1000 steps.
Compared to the previous relaxation calculation, the only difference in INCAR was that PREC was changed to Accurate and ENCUT was changed to values from 300 to 520,
and IBRION=-1 of course.

To solve such convergence problem, I also tried Gaussian Smearing with 1 kpoints or mixing parameters for molecules:

Code: Select all

AMIX = 0.2
BMIX = 0.0001
AMIX_MAG = 0.8
BMIX_MAG = 0.0001
But it still didn't converge.

Is there anything wrong in my inputs or what else can I try?
ALGO = 38? Manually set NGX ot NGXF?

Anything suggestions would helpful!!
<span class='smallblacktext'>[ Edited ]</span>

electronic convergence problem on molecule

Posted: Tue Oct 23, 2012 4:18 am
by tlchan
If it is an isolated molecule, then only the gamma point should be used for the calculation. You can use the other smearing schemes such that you can use a larger sigma for systems that are hard to converge.

electronic convergence problem on molecule

Posted: Wed Oct 31, 2012 11:19 am
by danith
Are you sure about ISPIN=1?

electronic convergence problem on molecule

Posted: Thu Nov 01, 2012 3:04 pm
by baliuzeger
Thank you both for your replies!

First, for the convergence problem, it's already solved by IALGO=38.

Second, for ISPIN=1 mentioned by danith,
do you mean that there may be localized spin moment within the molecule?
Or do the choice of ISPIN affect the convergence?